++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external TCP
Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

How can I modifiy this timeout. On Windows (winsock) the timeout is less
than one min with the same program.

I try to use setSock with SO_SNDTIMEO AND SO_RCVTIMEO (return 0 (no error)),
but it change nothing.

Did somebody have an idea to change this timeout?

Thanks

Serge
Well, assuming your client does a read() from the socket at a regular

On a sunny day (Fri, 26 Aug 2005 10:07:18 +0200) it happened "Serge
Wenger"
serge_wenger001@hotmail.com> wrote in <demijn$njs$1@atlas.ip-plus.net>:



++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external
TCP
Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

How can I modifiy this timeout. On Windows (winsock) the timeout is less
than one min with the same program.

I try to use setSock with SO_SNDTIMEO AND SO_RCVTIMEO (return 0 (no
error)),
but it change nothing.

Did somebody have an idea to change this timeout?

Thanks

Serge
Well, assuming your client does a read() from the socket at a regular
basis, test for an error return in read(), it will return EOF if the
server
shuts down.

You can also look for other errors and make a timeout loop:

time_t read_timer;
#define READ_TIMEOUT30 // seconds

/* read from socket */
while(1)
{
if(debug_flag)
{
fprintf(stderr, "before read(): socketfd=%d http_buffer=%p
http_bptr=%p\n",\
socketfd, http_buffer, http_bptr);
}

/* start the read timer */
read_timer = time(0);
while(1)
{
a = read(socketfd, http_bptr, content_length / 2);
if(debug_flag)
{
fprintf(stderr, "read() returned a=%d\n", a);
}

if(a > 0) break;
else if(a == 0) /* EOF server closed connection? */
{
fprintf(stderr, "read(): returned EOF (power failure, network,
interference?)\n");

report_read_timeout_flag = 1;
restart_flag = 1;
break;
}

if(debug_flag)
{
fprintf(stderr, "read() returned error because");
perror("");
}

if(errno == EAGAIN)
{
/* test for connect time out */
if( (time(0) - read_timer) > READ_TIMEOUT)
{
fprintf(stderr, "timeout in read(): (power failure, network,
interference?)\n");

report_read_timeout_flag = 1;
restart_flag = 1;
break;
}

usleep(1000);
continue;
}
else
{
fprintf(stderr, "fatal error in read() because\n");
perror("");

report_read_timeout_flag = 1;
restart_flag = 1;
break;
}
} /* end while read timer */
if(restart_flag) break;

Thanks for your suggestion, but it is exactly what I do. The problem is that

Serge
Yes, but if you use read you should use non-blocking!

On a sunny day (Fri, 26 Aug 2005 15:19:34 +0200) it happened "Serge
Wenger"
serge_wenger001@hotmail.com> wrote in <den4ta$p1j$1@atlas.ip-plus.net>:


++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external
TCP
Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

Yes, but if you use read you should use non-blocking!

"Jan Panteltje" <pNaonStpealmtje@yahoo.com> a écrit dans le message de news:
1125064942.bd240467b98fcd1a6b8d65d85f12e927@teranews...
On a sunny day (Fri, 26 Aug 2005 15:19:34 +0200) it happened "Serge
Wenger"
serge_wenger001@hotmail.com> wrote in <den4ta$p1j$1@atlas.ip-plus.net>:


++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external
TCP
Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

Yes, but if you use read you should use non-blocking!

Thanks for you answer, but I have the same problem qith non-blocking socket.

The server never send data, so I cannot use your method. The EOF arrive only
after 15 min. How can I change this timeout?

Help please.

Serge
Well, I dunno, but if the client only sends the data, and the server only listens,

I work on a TCP client (linux 2.4.25 elinos) that connect to
an external TCP Server (I can modifiy nothing on the TCP
server). The client send data to the server. If I switch off
the server, the TCP client see an error after more than 15
min.

How can I modifiy this timeout.

On Windows (winsock) the timeout is less than one min with the
same program.

I try to use setSock with SO_SNDTIMEO AND SO_RCVTIMEO (return
0 (no error)), but it change nothing.

Did somebody have an idea to change this timeout?

++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external
TCP Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

How can I modifiy this timeout. On Windows (winsock) the timeout is less
than one min with the same program.

I try to use setSock with SO_SNDTIMEO AND SO_RCVTIMEO (return 0 (no
error)), but it change nothing.

Did somebody have an idea to change this timeout?

On a sunny day (Mon, 29 Aug 2005 16:44:25 +0200) it happened "Serge
Wenger"
serge_wenger001@hotmail.com> wrote in <dev708$e2g$1@atlas.ip-plus.net>:

"Jan Panteltje" <pNaonStpealmtje@yahoo.com> a écrit dans le message de
news:
1125064942.bd240467b98fcd1a6b8d65d85f12e927@teranews...
On a sunny day (Fri, 26 Aug 2005 15:19:34 +0200) it happened "Serge
Wenger"
serge_wenger001@hotmail.com> wrote in <den4ta$p1j$1@atlas.ip-plus.net>:


++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external
TCP
Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

Yes, but if you use read you should use non-blocking!

Thanks for you answer, but I have the same problem qith non-blocking
socket.

The server never send data, so I cannot use your method. The EOF arrive
only
after 15 min. How can I change this timeout?

Help please.

Serge
Well, I dunno, but if the client only sends the data, and the server only
listens,
and this infrequent too, then why not use a procedure / protocol that goes
like:
start:
client tries to connect to server.
server accepts connection
client sends data
client closes connnection
wait until enough data
goto start

For example in case of a http server, and you uploading something, the
client
would likely close the connection after uploading.

"Serge Wenger" <serge_wenger001@hotmail.com> wrote in message
news:demijn$njs$1@atlas.ip-plus.net...
++++++++++++++ **************
+ TCP Server +-------<--------* TCP Client *
++++++++++++++ **************

Hello,
I work on a TCP client (linux 2.4.25 elinos) that connect to an external
TCP Server (I can modifiy nothing on the TCP server).
The client send data to the server. If I switch off the server, the TCP
client see an error after more than 15 min.

How can I modifiy this timeout. On Windows (winsock) the timeout is less
than one min with the same program.

I try to use setSock with SO_SNDTIMEO AND SO_RCVTIMEO (return 0 (no
error)), but it change nothing.

Did somebody have an idea to change this timeout?

Follow the protocol specification. I'm assuming, since you didn't write
the server and can't change it, that there's a protocol specification
you're following. Are you dealing with a well-known protocol?

15 minutes?!?? It should take 2 hours IF you've got the
KEEPALIVE option set. Without the KEEPALIVE option, it should
take forever.

You're confused.

The 15 mins he's seeing is probably the retransmission timeout eventually
expiring. See Stevens TCP/IP Illustrated Vol 1 (21.2). The client has sent
data. No response from the server has reached the client. TCP is
retransmitting the data using exponential backoff and it takes a while
before it gives up and errors. This should be around 9mins...

The 2 hr keepalive is only relevant if you don't send any data over the
connection (and if keepalive is turned on).

On Windows (winsock) the timeout is less than one min with the
same program.

Have you shutdown the server or just pulled out a plug or turned off a
router? I would have expected Windows to take around the same time to
discover the problem.

It's best to use a protocol level timeout...

15 minutes?!?? It should take 2 hours IF you've got the
KEEPALIVE option set. Without the KEEPALIVE option, it should
take forever.

You're confused.

The 15 mins he's seeing is probably the retransmission timeout
eventually expiring. See Stevens TCP/IP Illustrated Vol 1
(21.2). The client has sent data. No response from the server
has reached the client. TCP is retransmitting the data using
exponential backoff and it takes a while before it gives up
and errors. This should be around 9mins...

I think your are right. 15 seem common
(http://www.developerweb.net/sock-faq/flatfaq.php#faq14)

The 2 hr keepalive is only relevant if you don't send any data over the
connection (and if keepalive is turned on).

On Windows (winsock) the timeout is less than one min with the
same program.

Have you shutdown the server or just pulled out a plug or turned off a
router? I would have expected Windows to take around the same time to
discover the problem.

I switch off the server, not the router

It's best to use a protocol level timeout...

not possible in my case.

Thanks for you help. Now I am searching to modify the retransmission
timeout. It seem not so easy...

It's best to use a protocol level timeout...

not possible in my case.

Thanks for you help. Now I am searching to modify the retransmission
timeout. It seem not so easy...